uncertified haskell engineer | certified researcher
Gothenburg, Sweden
Joined January 2010
- Tweets28.1K
- Following2.1K
- Followers5.5K
- Likes167K
instead of solving fun programming puzzles, you’ll be finding the subtle architectural misstep in 10kn lines of agent-written code
The coding interview is dead, at least for my team at Anthropic. I've seen enough to know that it no longer provides signal. I removed it from loops this month.
I have a spicy, optimistic take - I think there's an opportunity for junior engineers: felixrieseberg.com/hunger-be…
it’s like if your phone had a built-in AI that was severely limited compared to what the models are doing
Windows Copilot was so terrible it might very well have seeded the AI opposition we now see among normies. Useless functionality, hyped to the moon, and a dumb key on every keyboard that did nothing worthwhile. youtube.com/watch?v=erRVsllM…
the spec to code pipeline is starting to feel very real
How do you _guarantee_ your agent's code is correct? You prove it.
Introducing Proof Driven Development Kit (PDD Kit) from @MSFTResearch.
PDD Kit helps turn your plan into a formal spec, then proves the agent's code meets that spec for every input. No vibe verification. PDD Kit is integrated with @GitHubNext's Chopin, so you can plan as a team.
PDD Kit takes an app's login flow from intent to a verified pull request:
It discusses your code, infers your intent as it goes, and proposes a formal spec in Verus, a proof language for Rust.
Once you approve the spec, the Verus verifier automatically proves the code meets it.
Where proof isn't possible, it falls back to empirical verification with tests and computer-use agents (CUAs).
It works on Rust today. No new language, no change in how you produce the underlying code, making formal methods more accessible.
The goal is a verified software factory, where engineers review specifications instead of code.
Come see it in action at GitHub Universe, Oct 28–29. Swing by the Speakeasy and ask for GitHub Next!
drug discovery is cool, but sadly verification is very expensive here. Not an NP problem at all
Over the past few months, Anthropic scientists have engaged in deep research partnerships with six organizations developing new medicines. We've also begun drug discovery efforts of our own.
In doing so, we’ve learned much more about how Claude can support the many stages of drug development.
In this blog post, we share some early results from these efforts, and describe how this work fits in with our wider ambitions in the life sciences.
anthropic.com/research/early…
the good thing about running a local model is that you don't have to have any blinking light or bell to know the turn is finished, you just wait till your fans spin down
i highly recommend reading the actual Frankenstein. It’s so much deeper than the movies make it seem
extremely relevant to the AI era, and eerie that the first sci-fi was about a man’s creation turning bad after being rejected by society / wanting to procreate
This is EVEN BIGGER
finally verified circuits!! And C!!
this will change hardware forever
Formally verify SystemVerilog and SVA (or autoforamlized properties).
Prove memory safety about your C programs.
Our AI theorem prover is now publicly available.
Try it out at harmonic.fun
There's a new version of this post
Wow, can't believe there is visible wear on the mechanism after using it ~40times/day for 5 years
the iPhone Duo broke after 75k folds
very disappointing to see for a ~ $2000 phone
Readers added context they thought people might want to know
During the test, the phone was exposed to dust and dirt, and just before it cracked, it spent 20 minutes in a bowl of ice.
Dust: youtu.be/YxcHtAooha4?t=…
Ice: youtu.be/gjUirIOVHNo?t=…
this is quite HUGE actually
not just the fact that it's been done, but also that it was done automatically
the future is bright!
my vision of the future: opus 5.5 or some equivalent model (useful for most), with the weights burned into an ASIC. Install it like a water heater in every house. tokens on demand for the masses
the best agent tasks are NP-complete
I feel like I need a word for the type of task that agents are really good at
I.e. where it has 100% of the data, it can run experiments, verify independently, and hill-climb towards its destination
On these tasks my trust level is totally maxed-out and the results are incredible
Just need a word for them
i think P=NP but the translation to P is so incredibly ugly (in our basis) that no self-respecting theoretical computer scientist has been willing to go down that path
i’ve always thought that mathematics was more of an art than science, so makes sense that they’re more like the artists than the scientists
Knowing @satnam6502, this is going to be HUGE. Exciting!
I don't think it's ugly, I think we're just working in some ugly basis. Like there will turn out to be a new constant, like e or pi (call it k), and all these algorithms have a k^... shape
Exciting!
But yeah complexity is so funny
like solving formulas that look like this (2-SAT):
(a ∨ b) ∧ (¬a ∨ c) ∧ (¬b ∨ ¬c)
linear time, trivial
but this (3-SAT):
(a ∨ b ∨ ¬c) ∧ (¬a ∨ c ∨ d) ∧ (b ∨ ¬c ∨ ¬d)
NP-complete! Exponential time! (better now though)
Announcing three major scientific discoveries, one each in mathematics (harmonic analysis), computer science (complexity theory), and physics (quantum field theory).
AI starting to help us improve our understanding of the universe! openai.com/index/building-sc…
